4 Stanford PhDs, breaking into the 100 billion track

Chinese entrepreneur

2024-07-30 22:31Published on the official account of China Entrepreneur magazine in Beijing

The full text is 3970 words, and it takes about 12 minutes to read

Highlight the point

A general-purpose robot developed by Dr. 014 Stanford is transforming food production lines to improve the taste and productivity of cod steaks.

02 The general-purpose robot adopts humanoid technology, mainly based on force control, supplemented by visual recognition, and has higher flexibility and adaptability.

03After 8 years of R&D accumulation, Feixi Technology has launched the world's first adaptive parallel robot and a new generation of general-purpose force-controlled grippers.

04 In addition, Feixi Technology also released the Qiongzhi AI system, which aims to empower the ecology of general robots and allow more people to create valuable things on this platform.

05At present, Feixi Technology's products have formed a certain benchmarking effect among leading enterprises in many industries, and their commercialization has increased by about 3 times every year.

The above content is generated by Tencent's hybrid model and is for reference only

Text: "Chinese Entrepreneur" reporter Zhao Dongshan

Editor|Li Wei

Image source: Respondents

When you eat a fluffy cod steak, you may not think about the robot power behind it.

In the past, in order to achieve the fluffy texture of the cod steaks, workers on the processing line had to pile the cod and breadcrumbs together and press them with a certain amount of force. To ensure food safety, workers also need to wear protective clothing and work hard in low temperatures. Today, four Stanford PhDs are transforming this product line with a general-purpose robot developed by Dr. Stanford. After blind testing, the cod steak produced by the robot tastes better, and the output efficiency is higher and more stable.

Wang Shiquan is one of the core founding members of the general robotics company. When he was an undergraduate at Zhejiang University, he chose robotics, which he believes is an interdisciplinary discipline that is more likely to burst out with sparks of innovation. After graduating from Zhejiang University, Wang Shiquan entered Stanford University to study for a doctorate, during which he joined the Bionic and Dexterous Operation Laboratory of Stanford University, and accumulated more than 10 years of experience in robot research and development.

In 2016, Wang Shiquan and a number of Stanford alumni founded Feixi Technology, along the path of humanoid technology innovation, focusing on the research and development and production of adaptive robots integrating industrial-grade force control, computer vision and artificial intelligence technology, providing general robot solutions and services for customers in different industries, and it is estimated that the proven application market size has reached 100 billion yuan.

In the past 8 years, Feixi Technology has completed a total of 4 rounds of financing, with investors including GSR Ventures, Gaorong Capital, Zhen Fund, Shunwei Capital, Magic Capital, Yunfeng Fund, China Merchants Capital and many other well-known institutions, as well as continuous support from Meituan, Pearl River Group, New Hope Group and other industrial parties, and has been reinvested by old shareholders for many times. According to public information, Feixi Technology's valuation after the B+ round of investment in 2022 will reach $1 billion, ranking among the unicorns.

In terms of commercialization, Feixi Technology's adaptive robots have entered the production processes and production lines such as precision assembly, screw tightening, surface coating, flexible grinding, flexible packaging, surgical operation, massage physiotherapy, laboratory automation, meat cutting, food processing, etc., covering automobiles, 3C, home appliances, FMCG, medical, biomedicine, agriculture and food.

difference

At the beginning of his business, Wang Shiquan chose a route that was completely different from the robots on the market, that is, to create a general-purpose robot through humanization.

The significant difference between it and other existing robots is that most of the existing robots use position control methods combined with additional visual or force sensors and a large number of non-standard equipment, and finally integrate into an intelligent solution, but in Wang Shiquan's view, such a technical framework restricts the flexibility and application scope of the robot.

Under such a technical framework, most of the robots are still serving some specific scenarios, or engaged in some structured work, and there are still a large number of tasks that can only be completed manually in major industries, challenging the product quality consistency, service quality and comprehensive cost of enterprises, especially in precision industries such as precision device plugging and assembly, complex surface grinding and polishing, as well as sorting and packaging, food processing and other heavy labor and poor working environment scenarios.

"Taking people as a reference, when people are doing things, the core concept is not precise trajectory memory, but touching and real-time adjustment of actions through force sensation. Therefore, the general-purpose robot we created needs to have the dexterous limb structure that humans have, and to complete the task in a way that focuses on hand adjustment and visual guidance, and can continuously comprehend and improve their own skills through muscle memory. Wang Shiquan told "Chinese Entrepreneur".

It has been fifty or sixty years since the development of general robots, but the penetration rate is still very low, Wang Shiquan believes that the humanoid path centered on Lijue may be a good breakthrough.

In addition, unlike many robot companies that focus on the exploration of the field of vision, Wang Shiquan believes that embodied intelligence is a real-world interaction, and if you want to make robots really do many things like people, in addition to vision, force control is more important.

"It is through the hands or the tools that are held, constantly acting with the environment, just like the interaction and learning with the whole environment after the birth of a baby, he can't just rely on his eyes to see, he also has to experience, he has to explore with both hands and even the whole body with the real environment, and project his own perception and action into a concrete world in a closed loop." Wang Shiquan explained.

In fact, even now, there are still very few people who have really practiced the theoretical route of force control at the global level, and it is in a state of non-consensus, but Wang Shiquan firmly believes that force control is a very necessary link to humanoid robots, and finally truly like people, or even beyond human capabilities.

accumulate

According to a report released by Morgan Stanley at the end of 2022, the global robot penetration rate is currently 0.9%, and robots account for about 1% of the labor market.

Wang Shiquan has personally experienced the evolution of robots in the past 15 years. In his opinion, whether it is industrial robots or collaborative robots, the development process is very long, and the price is very low, but the penetration rate is still very low. He believes that the most critical factor in this is that most robots on the market do not use tools and adapt to the real world as much as possible, but let the real world adapt to robots through certain modifications.

This is also his original intention to develop general-purpose robots. "The reason why the general robot is called universal must be that its products themselves can adapt to a variety of industries, without adding a lot of additional auxiliary equipment, on-site transformation and human adjustment."

In April 2019, at the Hannover Messe in Germany, Feixi Technology released its adaptive robot technology and demonstrated its first adaptive robotic arm, Rizon. After 5 years of iteration and optimization, Wang Shiquan said that Feixi Technology's robots have been able to adapt to the production line environment.

For example, for some conveyor lines with vibration, similar to human wiping and shaking tables, Feixi Technology adaptive robot no longer needs someone to press the table or incoming materials tightly, nor does it need high-precision vision sensors, completely with direct control of the contact force as the core, supplemented by low-cost visual judgment, can adapt to the environment and complete the work with quality and efficiency.

At Stanford University's Bionic and Dexterous Operations Laboratory, Wang Shiquan led and participated in the completion of projects such as rock climbing, underwater and home humanoid robots. Ye Xiyang and Zhong Shuyun, co-founders of Feixi Technology, are also Ph.D. and postdoctoral fellows in the Stanford University Robotics Laboratory, both of whom have more than 15 years of experience in robot software and hardware research and development; Lu Cewu, chief AI scientist of Feixi Technology, is a postdoctoral fellow in the AI Lab of Stanford University, a professor in the AI Lab of Shanghai Jiao Tong University, and a Changjiang Scholar.

In multiple laboratories at Stanford, Wang Shiquan and the core founding team have completed the technical reserves in the robot force control ontology and AI brain, which are the core parts of the general robot itself.

After more than 8 years of accumulation, the flywheel of Feixi Robot in technology, products and business is running rapidly. In 2021, the robot ontology developed by Feixi Technology will be basically mature; At the end of 2022, the core capabilities of Feixi Robot crossed the critical point of application value discovered by the previous team and began to harvest large-scale replication orders; At the end of 2023, benefiting from the maturity of product versatility, Feixi Robot broke the circle and started the accelerated development of cross-industry.

Wang Shiquan recalled with emotion: "It is indeed very long and difficult to walk this road, and it requires a lot of patience, but it is also very fulfilling. And one of the moments they would love to see in the process was a shift in customer attitudes.

At the beginning of contact, most customers will throw out the question of "why can't it be solved before, it can be solved now", until Wang Shiquan led the team to really overcome and prove it to them, customers will completely rethink their previous judgment, and even open more and more scenarios to the Feixi Technology team.

focus

Most robot entrepreneurs will first choose to do deep and thorough work in a certain field and then expand and try to do more applications, but Wang Shiquan has focused on the versatility of general robots from the beginning.

Wang Shiquan said that the focus of general robots is not the same as the focus of doing a certain industry segmentation solution, it is more from the industry backwards to understand the process behind the application of different scenarios, refining the commonality of the techniques behind the process itself, and focusing on common needs.

At the beginning of its establishment, the Feixi Technology team found that in the manufacturing sector, assembly and surface treatment actually have very common methods, based on which they can radiate to thousands of industry segments. Therefore, although the current Feixi technology seems to have a very diversified application scenarios and industry coverage, it can be condensed into a number of different "meta operations" that can be completed by the same Feixi robot, such as pressing, inserting, rolling, wiping, twisting, knocking and so on. These "meta-operations" only need to be set by different process parameters, or equipped with AI algorithms based on process understanding and visual judgment, to achieve different task purposes in a variety of production lines and application scenarios.

In the process of creating products, Wang Shiquan led the team to mature the robot ontology in the way of force control and visual recognition, breaking through the limits of various capabilities in a full-stack manner from sensors to drive units to the whole machine control, and finally optimizing the product with people as a reference system; On the other hand, Wang Shiquan went to look for scenarios where labor is intensive, the process cannot be optimized and iterated for a long time, and the customer wants to automate but there is still no mature solution.

Versatility determines the unique path to productization. Wang Shiquan said that if you go to blast an application point at a fixed point, become a big fish in a small pond, and then quickly go to other places, there are two natural challenges for the team:

First, for the introduction of each innovative application in the manufacturing industry, it generally needs to go through a very long process cycle to guide customers to try and verify. This is determined by the attributes of the industry, the degree of innovation and the business model, and cannot be accelerated by increasing investment or relying on manpower;

Second, the general environment and the situation of enterprises in various industries are constantly changing, and if the same strategy is used to pile up on an application point, the risk is very large.

In the past 8 years of R&D process, Wang Shiquan found that the performance of the parts on the market in many links was not enough, so from the torque sensor, joint controller, integrated joint, to the planning algorithm of the entire arm, to the operating system, etc., all links need to be self-developed by the team.

But it is precisely because of this that every step taken in the past has become a deep technical barrier for Feixi Technology. From explicit and implicit process patents, to algorithm barriers to the entire business model, Wang Shiquan led the team to produce their own iterations, develop their own applications, and finally feedback and iterate, and constantly optimize to form a flywheel effect.

future

At the beginning of 2023, Feixi Technology released the world's first adaptive parallel robot and a new generation of general-purpose force-controlled grippers, in addition to the noteworthy release of the Dome AI system.

Just like NVIDIA's CUDA computing platform based on its GPU and other hardware products, the Qiongzhi AI system released by Feixi Technology is also an infrastructure based on the uniqueness of the product, which aims to empower the ecology of general robots and allow more people to create valuable things on this platform.

Today, a set of robot ontology + software meta-operation and AI solutions of Feixi Technology, the price ranges from 100,000 yuan to 340,000 yuan, and it is forming a certain benchmarking effect after entering the leading enterprises in different industries.

Regarding the future business increment space, Wang Shiquan said: "In fact, the industry covered by Feixi Technology is already very broad, and the growth of more business volume in the future, in addition to sowing more seeds, will come from how to really make the seeds buried in each industry segment and the seedlings that break the ground become towering trees." ”

Looking back on this entrepreneurial journey, Wang Shiquan said that the first few years were like a period of enjoyment of technological exploration, and everyone had a clear general direction, with the mentality of "just do it", and devoted themselves to research and development technology. At the end of 2019, the robot products developed by the team were officially released, and the first batch of seeds were available.

But then there was a dark period, because it was a new product route and technical architecture, and there was no previous coordinate system as a reference, Wang Shiquan and his team didn't even know how good the core performance of the product had to be in order to meet the needs in the application process, and they could only continue to explore and break through the boundaries of various capabilities.

At the same time, on the one hand, Wang Shiquan and other management teams should grasp the rhythm of the company's operation and financing, and take into account the organization and management during the rapid expansion period of the company's team; On the other hand, it is necessary to manage the customer's expectations, so that the other party can patiently trust the team and go through the long application iteration period until the final validation is successful.

By the time the maturity of the product breaks through the critical point of real industrial landing, some core capabilities have even been improved hundreds of times compared with the time of release. With the first success story, the light finally dawned, the flywheel of technology, products, and commercialization continued to accelerate, and the company began to grow in the way expected.

This is the first time Wang Shiquan has led the entrepreneurship, and he has finally become handy, and the team and business have also begun to accelerate in a sprint.

In his opinion, the ideal state of the general robot will have three characteristics: first, it can complete all kinds of work with quality and quantity like a person, and even surpass people, which is the most core, but in fact, it is also the most difficult; second, it learns from the experience of masters in different industries faster and faster, and the cost is getting lower and lower; Third, robots with different skills are becoming more and more convenient and intelligent when interacting with people.

These, in the next 10 years, there will be more breakthroughs in Feixi Technology.

View original image 362K

• 

• 

•